Ask Cat › AI Tool Summary › Gemini
Vercel AI Gateway Shipped Three Models in September — but That 50% Off Ended on the 7th: Always Find the Deadline First
- Free tier:There is
- Cheapest paid plan:US$4.99/mo and up
- Free quota:NT$0/month, free to anyone with a Google account. Model available: …
- Last checked:2026-09-08
Article last updated:2026-09-08
Vercel AI Gateway shipped three things on 2026-09-02. One of them was a discount — and by our verification date, 2026-09-08, it was already over.
This is less about the release list than about a pattern that catches people out: AI platform promos often last days, and no one sends you a note when they lapse.
1. What landed on 2026-09-02
| Item | Detail | Deadline |
|---|---|---|
| Muse Spark 1.3 (Meta) | Callable through a unified API key, with spend tracking and request tracing | None (launch) |
| Gemini 3.8 Flash (Google) | Added to the gateway with consolidated API management and monitoring | None (launch) |
| GLM-5.3 at 50% off | Served by DigitalOcean, half price on AI Gateway | Through September 7 |
Vercel’s exact wording: “GLM-5.3 served by DigitalOcean is 50% off on Vercel AI Gateway through September 7.” Through September 7 means it ends there. Today is the 8th. That one is gone.
2. The promo that is still live
The same changelog, dated 2026-09-04: Ling 3.0 Flash Sante from Inclusion AI is free on Vercel AI Gateway through October 4, accessed via a promotional model ID.
That one is still valid at time of writing — the offer worth looking at today if you want to try a model without paying, rather than chasing an expired half-price deal.
Also on 09-04: GPT-6 Astra became available through the gateway, described as supporting “coding, computer use, research, and multi-step professional workflows.” A launch, not an offer.
3. Why “find the deadline” deserves to be a habit
This round happens to show both endings: one promo dead, one alive. The only difference is which day you read the news.
Three practical notes:
- When an AI story says “% off” or “free”, look for two dates — the publication date and the expiry date. If either is missing, assume it can stop at any time.
- Time-limited free models usually run on a different model ID (Ling’s promo does). When the promo ends, the usual outcome is not a higher bill — the ID simply stops working and your code breaks. That distinction matters, because it breaks at night.
- Put the expiry in a calendar, not in your memory — especially if a promotional model is serving production traffic.
4. What to actually do
Just trying models? Use the promo that is still running (Ling 3.0 Flash Sante, through 10-04). Do not spend time hunting an expired discount.
Building a product? Do not depend on promotional models. Use them to cut development-phase cost, but swap the model ID back to the standard one before launch, and confirm you can afford the standard rate.
Comparing platforms? The value of a “one key, many providers” gateway is not the occasional discount — it is switching models without changing architecture. Promos are the sweetener; portability is the reason.
Details on the Ling promo: Ling 3.0 Flash Sante free through October 4. Wiring one key to many models: AI Gateway setup. Plan comparison on the Gemini tool page.
Sources read directly on 2026-09-08: the 2026-09-02 and 2026-09-04 entries in the official Vercel changelog. The GLM-5.3 deadline (through September 7), Ling 3.0 Flash Sante free through 10-04, the Muse Spark 1.3 and Gemini 3.8 Flash launches and the GPT-6 Astra availability are all as published. We did not obtain post-discount effective rates, nor verify how promotional model IDs behave after a promo ends; point 2 in section 3 is general promo practice, not a Vercel statement.
Let's take a look at these
- Gemini Comprehensive Introduction: Pricing, Features, and Actual Limitations
- Gemini Is the free quota enough?
- Gemini Alternatives
- Comprehensive Free Quota List for All Tools
More verified articles on this tool
- Google Judged Thousands of AI Agents: the Top Entries All Used the Same Four Patterns, One Cut 40% of Model Calls
- Gemini Can Now Work Directly on Your Google Photos Library — Three Limits (US Only, English Only, Paid) and Three Safety Defaults, Verified 2026-09
- Google's WeatherNext 3 Is Live: Hourly Forecasts at 5km, and Four Ways Developers Can Pull the Data

